How to Audit vCenter SSO Accounts After a Breach
Security Intermediate 4 min read

How to Audit vCenter SSO Accounts After a Breach

Audit vCenter SSO accounts after a breach: list every vsphere.local account, find rogue admins, and rotate the keys an attacker could forge.

Published: November 21, 2026 • Updated: November 21, 2026
vcenterincident responsesingle sign-onaccount audit

Auditing vCenter SSO accounts after a compromise means listing every account and group in the vsphere.local domain, comparing them against what you created, and checking the appliance for access the attacker left behind. You do this because vCenter Single Sign-On accounts grant control over every virtual machine vCenter manages, and the ARMORY CASTLE campaign behind CVE-2026-123890 created rogue SSO administrators to keep that control. Patching closes the door. This audit tells you whether someone was already inside.

Why audit vCenter SSO accounts at all?

A vCenter Single Sign-On account can hold administrator@vsphere.local level rights, which means authority over hosts, clusters, virtual machines and the certificates that sign authentication tokens. An attacker who bypasses authentication does not need to keep exploiting the bug if they simply add themselves as an SSO administrator. That account survives the patch.

Mandiant observed exactly this in the CVE-2026-123890 intrusions: after exploiting the flaw, the group created a new SSO administrator account, in one case named with an sps prefix, and a backdoor machine account in the directory. A patch stops future exploitation of the vulnerability. It does not remove an account that already exists. That is why Broadcom's forensic triage flag and Mandiant's own guidance both say to patch and then investigate.

How do you list every SSO user and group?

Start in the vSphere Client under Administration, then Single Sign On, then Users and Groups. Select the Groups tab, confirm the vsphere.local domain is selected, and open the Administrators group. Every member there can administer the SSO domain itself. Compare the membership against your own record of who should be there.

The web client has a trap: it shows only the first 200 users or groups and will not display newly created entries unless you search for them. A rogue account created during an intrusion can sit past that cutoff. To get the complete list, Broadcom's guidance points to PowerCLI with the community SSO Admin module:

Install-Module VMware.vSphere.SsoAdmin
Connect-SsoAdminServer -Server vcenter.example.com
Get-SsoGroup -Domain "vsphere.local"
Get-SsoPersonUser -Domain "vsphere.local"

You can also confirm the built-in Administrators group membership directly on the appliance over SSH, using the directory CLI that ships with vCenter:

/usr/lib/vmware-vmafd/bin/dir-cli group list --name Administrators

Any account in that output that you cannot tie to a person or a known service is a finding. Note it, do not delete it yet, and preserve it for whoever runs the investigation.

What else should the audit cover?

Rogue accounts are one artifact. The CVE-2026-123890 activity touched several others, so widen the audit:

  • SSH state on the appliance. The attacker enabled SSH through the VAMI management API, used it, then turned it off. Check whether SSH is enabled when it should not be, and review the appliance's access and audit logs for the toggle and for logins.
  • The STS signing certificate. The exploit extracted the Security Token Service private signing key, which allows forged SAML tokens (a Golden SAML attack). If you confirm access, treat the signing key as stolen. Broadcom KB 318968 covers checking and replacing the STS signing certificate with the vCert tool.
  • Secrets on the appliance. The vCenter master encryption key and the PostgreSQL database password were both targeted. Rotate them if there is any sign of access.
  • ESXi and downstream hosts. The campaign pivoted from vCenter to ESXi, added an SSH key to the root account's authorized_keys, and mounted VMFS volumes to extract data. Check ESXi root authorized keys, look for hidden directories on datastores, and review vpxuser and root logins.

How do you tell a legitimate account from a rogue one?

Match every account against three things you control: your provisioning records, your change tickets, and your identity source. Service accounts and solution users have predictable names and were created during installation or integration work you can point to. A human administrator maps to a real person. Anything that matches none of those, especially an account created in the exposure window, is suspect.

For CVE-2026-123890 the exposure window opened in early September 2026, when Mandiant first saw exploitation, and ran until you patched. Focus the review on accounts and certificate changes in that range. If you find a rogue administrator, do not simply delete it and move on: forged SAML tokens signed with a stolen key will keep working until you rotate the signing certificate, so credential and key rotation has to follow the cleanup. The order matters, and it is the same lesson as in why you rebuild before rotating credentials after a breach.

This audit is step three of the response. Step one is patching, covered in our report on the exploited vCenter authentication bypass. Understanding why the directory channel carried so much trust is covered in what the Secure Remote Password protocol is.